Charnchai Pluempitiwiriyawej

Active Contours With Automatic Initialization For Myocardial Perfusion Analysis

Quantitative analysis of myocardial perfusion requires detection of myocardial boundaries in many short-axis MR images. Manual tracing of myocardial boundaries is a time-consuming and tedious task, which may limit the clinical use of quantitative analysis. In this paper, we propose an automatic detection algorithm based on active contours. For initialization, starting contours need to be defined. Instead of manually tracing the initial contours, our approach presents an automatic method for finding initial contours of epicardium and endocardium. Once the initial contours are defined, we apply the active contours called stochastic active contour scheme (STACS) for image segmentation

Automated tumour boundary delineation on 18F-FDG PET images using active contour coupled with shifted-optimal thresholding method

This study presents an automatic method to trace the boundary of the tumour in positron emission tomography (PET) images. It has been discovered that Otsus threshold value is biased when the within-class variances between the object and the background are significantly different. To solve the problem, a double-stage threshold search that minimizes the energy between the first Otsus threshold and the maximum intensity value is introduced. Such shifted-optimal thresholding is embedded into a region-based active contour so that both algorithms are performed consecutively. The efficiency of the method is validated using six sphere inserts (0.52-26.53 cc volume) of the IEC/2001 torso phantom. Both spheres and phantom were filled with (18)F solution with four source-to-background ratio (SBR) measurements of PET images. The results illustrate that the tumour volumes segmented by combined algorithm are of higher accuracy than the traditional active contour. The method had been clinically implemented in ten oesophageal cancer patients. The results are evaluated and compared with the manual tracing by an experienced radiation oncologist. The advantage of the algorithm is the reduced erroneous delineation that improves the precision and accuracy of PET tumour contouring. Moreover, the combined method is robust, independent of the SBR threshold-volume curves, and it does not require prior lesion size measurement.

Active contour using local region-scalable force with expandable kernel

In this paper, we propose a local region-scalable active contour with expandable kernel for image segmentation. We call it LREK active contour. Our model uses intensity values of pixels on a set of scalable kernels along evolving contour. These kernels are to direct contour front towards objects boundary within an image domain. Key feature of our model is that scale of the kernels increases gradually until the boundary is detected. So, our LREK may reach the boundary faster than some other methods. We compare performance of our LREK to existing region-based models that using local region descriptor. Experimental results show more desirable segmentation outcomes of our method. Our LREK performs effectively in segmenting noisy, concave boundary, non-uniform, and heterogeneous textures objects with a large capture range and fast convergence. Moreover, our Gaussian LREK is able to trace blur or smooth boundary.

LEFT VENTRICULAR SEGMENTATION USING DOUBLE REGION-BASED SNAKES

We present a double region-based snake method to segment endocardial and epicardial boundaries of the left ventricle in cardiac MR images. The method is based on a region-based parametric active contour. The first active contour is designed to segment the endocardial boundary while the second active contour is to segment the epicardial boundary. We insert the inter-contour forces between the two convolving contours in order to control the distance between them. Our proposed method not only has an ability to segment the endocardial and the epicardial boundaries of the left ventricle simultaneously without requiring any training sets, it is also able to segment cardiac MR images with low contrast between the myocardium and other organs. We initially place both contours within the epicardial boundary of the left ventricle. Furthermore, the problem that one of the contours may, on occasions, be attracted to the boundary of the papillary muscles is solved by checking for the convex hull condition. Our results show successful simultaneous segmentations of the endocardial and the epicardial boundaries, thus the left ventricle of the heart in each MR image

Feature-based motion detection and tracking on approximate 3D ground plane

The success of movement detection based on the distance moved in a 2D image sequence depends highly on the angle between a cameras optical axis and the normal vector of the ground plane on which the moving object is traveling. When the same 3D displacement occurs at various positions in the scene, the higher the angle is, the greater the distance observed in the image at a position close to camera differs from (strictly speaking, is larger than) those happening at the far end. As a consequence, a detection failure and/or false alarm may occur if no such 3D geometry is utilized. This paper estimates a 3D ground plane, which is then used to measure the approximate 3D displacement of features being detected and tracked. The 3D distances are therefore available and utilized in deciding whether they are of moving objects or just blinking features caused by illumination changes. FAST points are used to enhance a real-time system. Experimental results show superior performance in tracking: a longer trace of continuous tracking, a higher number of detected moving features, earlier detection, better recall rate, no misses, and no false alarms. A SURF descriptor and FLANN matcher were utilized here, however the robustness was not much enhanced when compared to the expense of finding the best match.

Edge type-selectable active contour using local regional information on extendable search lines

In this paper, we propose a novel active contour method that selects the desirable object based on its edge type. Our method is an extension of the local regional active contour with extendable search lines (LRES) [1]. However, we have added to it an ability to look for a particular edge type. To search for the object boundary, the method uses the intensity profile along search lines that are normal to the contour front. The length of these search lines is gradually extended until the object boundary is found. In this paper, we utilize the sign of the difference between the intensity profiles along the search line that are inside and outside the contour as a switching parameter in order to manage the forces that are to drive the contour toward the object with desirable edge type. With the same initial contour, our novel active contour can move toward different objects in the image by setting the edge type parameter. We compare the performance of our edge type selective LRES method to other existing selectable edge type active contour. The results show that our method provides more desirable segmentation outcomes, particularly on some images where other method may fail. Not only is our method robust to noise and able to reach into a deep concave shape, it also has a large capture range and performs well in selecting the object of desirable edge type within the image with the same initial contour.

Comparison of dense matching algorithms in noisy image

In this paper, we compare two correlation techniques for dense matching used in image corresponding problem, namely, the Sum of Squared Difference (SSD) and Normalized Cross Correlation (NCC). Both algorithms look for part of the image that matches a template based on intensity information. The window of the template is of Voronoi size, according to each Voronoi cells. The corresponding seed relations in each cell until all pixels within each cell are processed using SSD and NCC algorithms. In our experiments compare the performance of SSD and NCC in image with additive Gaussian noise, salt and pepper noise, and speckle noise. We found that SSD is more robust to noise than NCC in all cases.

3D left ventricular segmentation using double active contours and double active surfaces

We propose a 3D left ventricular segmentation method for cardiac MR images. There are two steps in our method. In the first step, we improve our double active contours for segmenting the endocardial and the epicardial boundaries simultaneously without requiring any training sets. In the second step, we designed a double active surface method to segment the volume within the endocardium and the thickness of the myocardium simultaneously from a set of 2D contours generated from the first step. Our results show successful simultaneous segmentation of the 3D left ventricular model of the heart.